A spectral layer model in speech and music perception

Abstract

According to a model of the peripheral hearing system establishing a masking and overmasking paradigm, speech and music signals have been split into several spectral layers by means of spectral subtraction. Spectral components below the so-called psychoacoustical irrelevance threshold are eliminated in a first processing step. The resulting signal, which exhibits no noticeable difference to the unprocessed one, is fed into an overmasking procedure applying flattened hearing masking functions. This action implements a kind of a spectral peak detector device. Consequently, the discrimination between spectral peaks (auditory figure) and background components, especially in reverberant environments, is improved. As a result, different audible signal parts according to different spectral layers are obtained, separating spectral peaks and weaker spectral components. In a certain range, both parts of speech and music signals are equally intelligible. In music, leading voices can be extracted and separated from the orchestra sound. Recent research indicates that this type of figure-background discrimination based on the dynamic characteristics of the acoustic signal only can exhibit several perceptually relevant features in listeners with normal hearing as well as in subjects with sensineural hearing impairment.